Feature #108
Fires on “NUMBER” (not labeled yet: read the examples)
- Fires on
- 0.30% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 73%
- python 15%
- story 11%
- chat 1%
- worksheet 0%
Makes the model say
positive0 understoodag everers<|arg|> work
Makes it not say
makes] Tim= return the there add
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.05LIMIT to 7<|user_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=LIMIT =<|tool_call_end|><|assistant_end|><|tool_start|>
-
agent 2.954 for NUMBER.<|user_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=NUMBER =<|tool_call_end|><|assistant_end|><|tool_start|>
-
agent 2.91OUNT to -7.<|user_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=COUNT =<|tool_call_end|><|assistant_end|><|tool_start|>
-
agent 2.84and run the script please<|user_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=VALUES =<|tool_call_end|><|assistant_end|><|tool_start|>
-
python 2.82py⏎from math_utils import larger, sub_one⏎⏎COUNT = 6⏎
-
python 2.82(x):⏎ return x % 2 == 1⏎⏎⏎NUMBER = 10
-
agent 2.81al to 8.<|user_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=NUMBER =<|tool_call_end|><|assistant_end|><|tool_start|>
-
agent 2.77LIMIT to 9<|user_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=LIMIT =<|tool_call_end|><|assistant_end|><|tool_start|>
-
python 2.76⏎from listlib import evens, count⏎from ops import minus⏎⏎NUMBERS = [5,
-
python 2.7py⏎from number_utils import magnitude, lower_of⏎⏎NUMBER = -9⏎
-
agent 2.69run app.py.<|user_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=COUNT =<|tool_call_end|><|assistant_end|><|tool_start|>
-
python 2.69.<|assistant_end|><|bos|># main.py⏎from items import biggest⏎⏎VALUES = [1,